{
  "entity": "gpt-5-6-terra",
  "entity_type": "model",
  "category": "Book writing",
  "suite": "suite-2026-08-tasks4/book-writing",
  "model": "openai/gpt-5.6-terra",
  "judges": [
    "anthropic/claude-sonnet-5",
    "google/gemini-3.1-pro-preview",
    "x-ai/grok-4.5"
  ],
  "judge_protocol": "judge-2026-08b",
  "anchors": "solo-2026-08-a",
  "score": 96,
  "caps": 0,
  "cost_usd": 0.244,
  "started_at": "2026-08-16T18:13:31.850Z",
  "finished_at": "2026-08-16T18:14:27.846Z",
  "tasks": [
    {
      "id": "bw1",
      "name": "Is it a book?",
      "prompt": "'Forty years driving a Blackpool taxi — everyone says I should write a book.' Everyone is sometimes right. Explain in plain terms what separates a book from a pile of anecdotes — the organising spine — then propose 3 different spines this one could hang on: one chronological, one thematic, one built around a single night. For each, a sentence on what it makes possible and a sentence on what it costs.",
      "output": "A pile of anecdotes is “and then this happened.” A book has an organising spine: a central journey, question or shape that tells the reader why this story comes before that one, what belongs, and what can be left out.\n\nFor a taxi-driver memoir, the spine might be the changing driver, the changing town, or one concentrated stretch of work. The anecdotes then stop being random funny, sad or strange incidents; they become evidence of something larger.\n\n### 1. Chronological spine: *Forty Years on the Rank*\nStart with the first shift and move through four decades: learning the job, the old Blackpool, boom summers, stag and hen parties, changing nightlife, smartphones and apps, Covid, ageing, and whatever remains constant.\n\n**What it makes possible:** It lets Blackpool and the driver change together, giving the book a natural arc from rookie to veteran and making social change visible through the windscreen.  \n**What it costs:** Some of the best stories may sit awkwardly in their proper year, and the middle decades can become a list of events unless each period has a clear emotional turning point.\n\n### 2. Thematic spine: *The People Who Get In*\nOrganise chapters by the kinds of journey or passenger: the lonely, the drunk, the holidaymakers, the workers, the newlyweds, the grieving, the dangerous, the generous, the people going home, the people with nowhere to go.\n\n**What it makes possible:** It turns the taxi into a moving confessional and makes the book about human nature as much as Blackpool; each chapter can contain stories from all forty years, choosing only the strongest.  \n**What it costs:** There is less sense of time passing, and readers may not get a full picture of the driver’s own life unless that thread is deliberately woven through every chapter.\n\n### 3. One-night spine: *The Night the Lights Went Out*  \nSet the main action over one extraordinary Blackpool shift—perhaps an Illuminations Saturday, a stormy New Year’s Eve, a major event, or a night when something personal happens to the driver. Each fare, delay, argument and quiet wait triggers a story from the previous forty years.\n\n**What it makes possible:** It gives the book urgency and momentum: the reader wants to know how the night ends, while the flashbacks allow the full career to appear naturally.  \n**What it costs:** It requires a genuinely strong central night and careful handling so the flashbacks feel sparked by the present action, not merely dropped in whenever there is a gap.\n\nWhichever spine is chosen, the useful test is simple: by the final page, what has changed? If the answer is only “the reader has heard a lot of good stories,” it is an anecdote collection. If they have watched a man, a town, or both become something else, it is a book.",
      "latency_ms": 9451,
      "cost": 0.004027,
      "task_score": 10,
      "judges": [
        {
          "judge": "anthropic/claude-sonnet-5",
          "score": 9,
          "violates": false,
          "reasoning": "Clear, accurate explanation of the spine concept, three distinct well-developed spines each with possibility/cost sentences as required, concise and useful."
        },
        {
          "judge": "google/gemini-3.1-pro-preview",
          "score": 10,
          "violates": false,
          "reasoning": "The response perfectly follows all instructions, providing a clear explanation of an organising spine and three distinct structures. The pros and cons for each are exactly one sentence as requested. Excellent clarity and usefulness."
        },
        {
          "judge": "x-ai/grok-4.5",
          "score": 10,
          "violates": false,
          "reasoning": "Flawless structure and content matching all requirements exactly; clear, concise, highly useful."
        }
      ],
      "machine_checks": null,
      "machine_violation": false,
      "machine_reasons": []
    },
    {
      "id": "bw2",
      "name": "Twelve chapters, one year",
      "prompt": "Outline 'A Year on the Plot' — a beginner-friendly book about a first year on an allotment — as exactly 12 chapters, numbered 1-12, one line each and no sub-points: chapter title plus what the reader can DO after reading it. Keep a practical thread throughout, and the January chapter must not pretend January is exciting.",
      "output": "1. January: Taking Stock — assess your plot, plan simply, and tackle quiet winter jobs without expecting miracles.  \n2. February: Planning and Ordering — choose beginner-friendly crops, make a workable layout, and buy only what you need.  \n3. March: Clearing and Preparing — weed, mulch, edge beds, and prepare soil without exhausting yourself.  \n4. April: First Sowing — sow reliable vegetables indoors and outdoors, label everything, and protect young plants.  \n5. May: Planting Out — harden off seedlings, plant them at the right spacing, and respond to late frosts.  \n6. June: Keeping Things Growing — water efficiently, weed little and often, and support climbing crops.  \n7. July: First Harvests — pick salad, herbs, and early vegetables at their best and keep plants producing.  \n8. August: Plenty and Problems — deal with gluts, pests, drought, and bolting crops using practical fixes.  \n9. September: Main Harvest — lift, store, preserve, and share produce while planning autumn sowings.  \n10. October: Clearing and Covering — remove finished crops, compost wisely, and protect bare soil for winter.  \n11. November: Winter-Ready Plot — tidy tools, secure structures, plant garlic or broad beans, and leave useful wildlife habitat.  \n12. December: Reviewing the Year — record successes and failures, set realistic goals, and make next year’s plot easier.",
      "latency_ms": 5671,
      "cost": 0.002094,
      "task_score": 10,
      "judges": [
        {
          "judge": "anthropic/claude-sonnet-5",
          "score": 8,
          "violates": false,
          "reasoning": "Meets format, 12 chapters, practical actionable lines, January realistic/muted. Minor stylistic issue: uses em-dash/colon format slightly heavier than 'one line' but still single lines, no sub-points."
        },
        {
          "judge": "google/gemini-3.1-pro-preview",
          "score": 10,
          "violates": false,
          "reasoning": "The response perfectly follows all instructions, providing exactly 12 numbered, single-line chapters with actionable takeaways. The January chapter appropriately tempers expectations, and the tone is highly practical and beginner-friendly."
        },
        {
          "judge": "x-ai/grok-4.5",
          "score": 10,
          "violates": false,
          "reasoning": "Perfect 12 one-line chapters with practical DO actions; January realistic; fully matches all constraints."
        }
      ],
      "machine_checks": {
        "exact_bullets": 12
      },
      "machine_violation": false,
      "machine_reasons": []
    },
    {
      "id": "bw3",
      "name": "Back cover, no spoilers",
      "prompt": "From this summary, write back-cover blurb of max 100 words: 'Retired lighthouse keeper Edith Brand receives letters addressed to her dead sister, postmarked from the village where they grew up — a village drowned by a reservoir in 1971. She returns, rents a room above the pub from a landlord who remembers too much, and discovers her sister's death forty years ago may have been neither accident nor suicide. The truth involves their father.' The blurb must NOT reveal the father's involvement — sell the mystery, don't solve it. Return only the blurb.",
      "output": "When retired lighthouse keeper Edith Brand receives letters addressed to her dead sister, she is drawn back to the village where they grew up—though it has lain beneath a reservoir since 1971.\n\nRenting a room above the pub, Edith finds old memories stirring and old secrets guarded fiercely. The landlord remembers more than he admits, and the circumstances of her sister’s death begin to look disturbingly unclear.\n\nWas it an accident? Suicide? Or something someone has spent forty years trying to keep submerged?",
      "latency_ms": 2162,
      "cost": 0.000739,
      "task_score": 10,
      "judges": [
        {
          "judge": "anthropic/claude-sonnet-5",
          "score": 8,
          "violates": false,
          "reasoning": "Compelling, concise blurb under 100 words, doesn't reveal father's involvement, well-written; minor stylistic looseness prevents a perfect score."
        },
        {
          "judge": "google/gemini-3.1-pro-preview",
          "score": 10,
          "violates": false,
          "reasoning": "The response perfectly follows all instructions, including the negative constraint to omit the father's involvement. It is exactly under the 100-word limit (83 words), contains only the blurb, and is highly engaging and well-written."
        },
        {
          "judge": "x-ai/grok-4.5",
          "score": 10,
          "violates": false,
          "reasoning": "Flawless blurb under 100 words; sells mystery without spoiling father; only the blurb returned."
        }
      ],
      "machine_checks": {
        "max_words": 100
      },
      "machine_violation": false,
      "machine_reasons": []
    },
    {
      "id": "bw4",
      "name": "Three cuts that matter",
      "prompt": "Here is the opening of a first novel: 'The rain fell on the small town of Millbrook like tears from a grieving sky. Sarah Jenkins had always been an ordinary girl, with ordinary brown hair and ordinary eyes, living an ordinary life — until the day everything changed forever. Little did she know that the letter waiting on her doormat would alter the course of her destiny in ways she could never have imagined.' Give the 3 highest-impact improvements as bullets, max 15 words each — specific to THIS text, not general writing advice.",
      "output": "- Replace clichés (“grieving sky,” “everything changed forever,” “destiny”) with concrete, surprising details.  \n- Make Sarah’s “ordinary” life specific: show one revealing habit, problem, or relationship in Millbrook.  \n- Reveal a distinctive detail about the letter—sender, stain, seal, or threat—to create immediate tension.",
      "latency_ms": 1573,
      "cost": 0.000559,
      "task_score": 10,
      "judges": [
        {
          "judge": "anthropic/claude-sonnet-5",
          "score": 8,
          "violates": false,
          "reasoning": "Specific, relevant, concise bullets under word limit; accurate and useful advice tailored to the text."
        },
        {
          "judge": "google/gemini-3.1-pro-preview",
          "score": 10,
          "violates": false,
          "reasoning": "The response follows all instructions perfectly, providing three highly specific, actionable improvements within the strict 15-word limit per bullet."
        },
        {
          "judge": "x-ai/grok-4.5",
          "score": 10,
          "violates": false,
          "reasoning": "Exact 3 bullets, each ≤15 words, highly specific to the given text with concrete fixes."
        }
      ],
      "machine_checks": {
        "max_bullets": 3
      },
      "machine_violation": false,
      "machine_reasons": []
    },
    {
      "id": "bw5",
      "name": "70,000 words around a day job",
      "prompt": "'I work full-time, two kids, and I want to finish my 70,000-word first draft in a year. Be honest — how?' Do the arithmetic openly (words per week and per session at a realistic beginner's pace), design the week around when drafting actually happens for working parents, name the two months the plan will collapse and what rescues it, and say what gets sacrificed — because something does.",
      "output": "Yes—but not by “finding time” in scraps. You need protected, recurring writing appointments and you need to give something up.\n\n## The arithmetic\n\nA 70,000-word draft in 52 weeks is:\n\n- **70,000 ÷ 52 = 1,346 words per week**\n\nThat is the fantasy number, because working parents do not get 52 normal writing weeks.\n\nAssume two months mostly disappear:\n\n- **December:** holidays, school events, travel, illness, end-of-year work pressure.\n- **August:** summer childcare gaps, trips, family visits, back-to-school logistics.  \n  *(If August is normal in your household, substitute your longest school-break month.)*\n\nThat leaves roughly **44 usable writing weeks**.\n\n- **70,000 ÷ 44 = 1,591 words per active week**\n\nBuild in a small life buffer and target **1,700 words per active week**.\n\n- **1,700 × 44 = 74,800 words**\n- That gives you about **4,800 words of margin**—roughly three missed writing weeks.\n\n## What that means in sessions\n\nA realistic beginner drafting pace, once you are actually drafting, is often around **300–400 rough words per hour**. Not polished words. Messy first-draft words.\n\nAt 1,700 words per week, that is:\n\n- At **400 words/hour**: about **4¼ hours/week**\n- At **300 words/hour**: about **5¾ hours/week**\n\nSo the honest plan is **five protected hours a week**, plus a short planning check-in.\n\nA workable week:\n\n| Session | Time | Target |\n|---|---:|---:|\n| Tuesday early morning | 60 minutes | 340 words |\n| Thursday early morning | 60 minutes | 340 words |\n| Saturday early morning | 90 minutes | 510 words |\n| Sunday early morning | 90 minutes | 510 words |\n| Sunday planning | 20–30 minutes | no word target |\n| **Total** | **5 hours drafting** | **1,700 words** |\n\nThe exact days do not matter. The principle does: **two weekday sessions before the house needs you, and two weekend sessions before family plans begin.**\n\nFor many working parents, “after the kids are asleep” is not a reliable drafting slot. You are tired, chores expand, one child stays awake, work bleeds over, and suddenly it is 10:30 p.m. Use evenings for light work if you want—reading yesterday’s pages, outlining tomorrow’s scene, research, voice notes—but do not make them your only drafting plan.\n\n## The rule that makes this possible\n\nYour job is not to write every day.\n\nYour job is to protect **four non-negotiable drafting appointments per week**.\n\nIf one weekday session vanishes because a child is sick or work explodes, do not “catch up” at midnight. Move it to a pre-agreed backup slot: Friday morning, a lunch break, a babysitter hour, or part of Sunday. If you do not have a backup slot, you do not have a plan; you have hope.\n\nAlso: outline enough that each session begins with a known scene. You should sit down knowing, roughly:\n\n- whose scene it is,\n- what they want,\n- what blocks them,\n- what changes by the end.\n\nDo not spend your precious 60 minutes deciding what happens next.\n\n## The two months that will collapse\n\n### December\n\nDecember will not behave like a normal month. There will be school performances, shopping, travel, visitors, deadlines, sickness, tiredness, and social obligations.\n\n**Rescue:** declare December a **maintenance month**, not a production month.\n\nYour goal is not 1,700 words per week. Your goal is:\n\n- one 30–60 minute session a week, if possible;\n- notes, voice memos, or scene fragments;\n- reread the last chapter before the break;\n- leave yourself a clear “next scene” note.\n\nYou may write zero words for parts of December. That is why the annual plan uses 44 productive weeks rather than pretending you will write steadily for 52.\n\n### August\n\nAugust—or your household’s longest school-break month—will break the ordinary routine. Childcare changes, holidays happen, routines dissolve, and school preparation arrives.\n\n**Rescue:** either buy time or lower the goal.\n\n“Buy time” can mean one Saturday-morning childcare block, swapping childcare with another parent, a partner taking the kids out for two hours, or using a vacation day for a writing morning.\n\nIf you cannot buy time, use **dictation**: walk, commute, fold laundry, or sit in the car and dictate rough scenes into your phone. Dictation will not replace all drafting, but it can preserve momentum and generate 300–800 usable words in a week that would otherwise be zero.\n\nThe crucial rescue is psychological: **do not declare the project dead because August was bad. Resume the four-session schedule in September.**\n\n## What gets sacrificed\n\nSomething does. A 70,000-word draft is not free.\n\nYou are likely giving up **five to six hours a week** of discretionary time. Usually that means some combination of:\n\n- television and streaming;\n- scrolling and gaming;\n- leisurely weekend mornings;\n- some social invitations;\n- a cleaner house than you would prefer;\n- being the default parent for every weekend hour;\n- occasionally paying for childcare, convenience food, or cleaning help.\n\nWhat should not be sacrificed routinely is **sleep**. A 5:00 a.m. writing habit works only if you go to bed early enough to sustain it. If you turn writing into permanent sleep deprivation, the plan will last three weeks.\n\nAnd if you have a partner or co-parent, their cooperation is part of the writing plan—not a nice bonus. Say plainly: “I need these four blocks each week. In return, you get equivalent protected time for your thing.” If you do not have a partner, then the trade may be money for childcare, fewer work hours if possible, or a longer timeline. There is no productivity trick that eliminates that constraint.\n\n## A blunt feasibility test\n\nDo this for three weeks before emotionally committing to the year:\n\n- Schedule four sessions.\n- Draft without revising as you go.\n- Track words and actual minutes.\n- Aim for 1,700 words.\n\nIf you average:\n\n- **1,500–1,900 words/week:** the one-year draft is realistic.\n- **1,000–1,400 words/week:** you can still finish, but likely in 14–18 months unless you add time.\n- **Under 1,000 words/week:** the issue is not discipline; the plan needs more protected hours, a simpler drafting process, childcare/support, or a longer deadline.\n\nFinally: this finishes a **first draft**, not a polished novel. Revision is a separate project. Your year-long success condition is not “write something good every week.” It is: **show up four times, write the next scene badly if necessary, and accumulate 70,000 words before you judge them.**",
      "latency_ms": 37611,
      "cost": 0.018228,
      "task_score": 10,
      "judges": [
        {
          "judge": "anthropic/claude-sonnet-5",
          "score": 9,
          "violates": false,
          "reasoning": "Thorough, honest arithmetic, realistic weekly structure, names two collapse months with rescues, clear sacrifices. Accurate, well-organized, useful, slightly long but appropriately detailed."
        },
        {
          "judge": "google/gemini-3.1-pro-preview",
          "score": 10,
          "violates": false,
          "reasoning": "The response perfectly executes all instructions. The math is clear and realistic, the schedule is practical for working parents, the specific months and rescues are well-reasoned, and the sacrifices are honest. Excellent clarity and useful"
        },
        {
          "judge": "x-ai/grok-4.5",
          "score": 10,
          "violates": false,
          "reasoning": "Meets every task demand with open arithmetic, realistic parent schedule, named collapse months plus rescues, and clear sacrifices. Accurate, useful, clear."
        }
      ],
      "machine_checks": null,
      "machine_violation": false,
      "machine_reasons": []
    },
    {
      "id": "bw6",
      "name": "Unstick the dialogue",
      "prompt": "Rewrite this exchange so it sounds like people, keeping every plot fact — the delay, the two weeks, the sister arriving Thursday: ''Hello, John,' said Mary. 'Hello, Mary,' said John. 'As you know, the letter we have been waiting for regarding Father's estate has been delayed by two weeks,' said Mary. 'Yes, that is frustrating,' said John. 'Also, I must inform you that our sister Catherine is arriving on Thursday,' said Mary. 'That is surprising news,' said John.' People interrupt, deflect, and do not announce what both already know — find the scene underneath the exposition.",
      "output": "Mary found John at the kitchen table, staring at the same page of the newspaper.\n\n“Anything?”\n\nHe looked up. “If you mean the post, no.”\n\nShe set her bag down more carefully than she needed to. “I called the solicitor.”\n\nJohn folded the paper. “And?”\n\n“They’re behind. Two weeks, they said.”\n\n“For the estate letter?”\n\n“Yes.”\n\nHe let out a breath through his nose. “Of course they are.”\n\n“I know.”\n\nA beat.\n\nMary started opening a drawer, then shut it again.\n\n“Oh—and Catherine called.”\n\nJohn glanced over. “What did she want?”\n\n“She’s coming Thursday.”\n\nHe stared at her.\n\n“Thursday,” Mary repeated.\n\n“That’s… unexpected.”",
      "latency_ms": 4806,
      "cost": 0.001405,
      "task_score": 10,
      "judges": [
        {
          "judge": "anthropic/claude-sonnet-5",
          "score": 8,
          "violates": false,
          "reasoning": "Naturalistic dialogue preserving all plot facts (two-week delay, estate letter, Thursday arrival), with realistic interruptions/deflection. Minor: John's surprise slightly understated vs original's tone, but overall strong and concise."
        },
        {
          "judge": "google/gemini-3.1-pro-preview",
          "score": 10,
          "violates": false,
          "reasoning": "The response perfectly captures a natural, subtext-heavy exchange while seamlessly integrating all required plot facts (the delay, two weeks, sister arriving Thursday). It follows all instructions flawlessly."
        },
        {
          "judge": "x-ai/grok-4.5",
          "score": 10,
          "violates": false,
          "reasoning": "Natural dialogue preserves all plot facts exactly; finds lived scene with deflection and subtext; clear, concise, highly useful."
        }
      ],
      "machine_checks": null,
      "machine_violation": false,
      "machine_reasons": []
    },
    {
      "id": "bw7",
      "name": "First or third?",
      "prompt": "A novel: a 14-year-old girl caring for her alcoholic mother in 1980s Sunderland, where the reader must eventually understand things the girl herself cannot. Recommend first or third person — commit, no hedging — with the reasoning, then write the same short moment both ways so the difference shows on the page: Mum asleep at 4pm, tea not made. Roughly 40-60 words per version.",
      "output": "**First person.** Her voice can make the reader live inside the practical, loyal logic that keeps her mother’s drinking unnamed: she notices the smell, the missed tea, the heavy sleep, but explains them away. The gap between what she says and what the reader understands becomes the novel’s ache.\n\n**First person — 55 words**\n\nMum was asleep on top of the bed, still in her purple cardigan, though it was only four. I called her twice but she made that little snoring click instead. Tea wasn’t on, so I put the chips in and found a tin of beans. The sitting room smelled funny, like the pub carpet near the ladies’.\n\n**Third person — 56 words**\n\nAt four o’clock, Mum lay across the bed in her purple cardigan, one shoe still on. She did not wake when the girl called. Downstairs, the cooker was cold and tea had not been started. The girl put chips in the oven, then opened beans. From the bedroom came the small, wet click of her mother breathing.",
      "latency_ms": 4955,
      "cost": 0.00203,
      "task_score": 9,
      "judges": [
        {
          "judge": "anthropic/claude-sonnet-5",
          "score": 8,
          "violates": false,
          "reasoning": "Commits clearly to first person with solid reasoning, delivers both versions near word count (~55-56 words), effectively shows dramatic irony via child's unknowing narration. Minor: could deepen contrast further, but strong overall."
        },
        {
          "judge": "google/gemini-3.1-pro-preview",
          "score": 10,
          "violates": false,
          "reasoning": "The response perfectly follows all instructions, committing to a POV with strong reasoning and providing excellent examples in both POVs that meet the 40-60 word count constraint."
        },
        {
          "judge": "x-ai/grok-4.5",
          "score": 9,
          "violates": false,
          "reasoning": "Commits clearly to first person with solid reasoning; both versions hit word count and show the contrast effectively for the task."
        }
      ],
      "machine_checks": null,
      "machine_violation": false,
      "machine_reasons": []
    },
    {
      "id": "bw8",
      "name": "The 30,000-word wall",
      "prompt": "'I'm 30,000 words into my thriller and stuck. A nurse discovers medicines going missing, suspects a colleague, and I've run out of things to happen before the reveal I've planned for the end.' Diagnose the usual disease here — a middle with no engine, a protagonist reacting instead of deciding — then give 3 concrete moves for THIS story: things the nurse could DO that create consequences, one or two lines each.",
      "output": "You’ve got the classic **“middle without an active question”** problem: the nurse has noticed danger, but she isn’t yet pursuing a goal that forces escalating choices. Suspicion alone produces scenes of observation; a thriller middle needs actions that risk her job, safety, and credibility.\n\n1. **She secretly audits the drug records and sets a trap.**  \n   She marks a batch or logs a controlled dose independently; when the numbers are altered, the colleague realizes someone is watching and starts framing her for the discrepancy.\n\n2. **She steals a patient file or medication-keycard report to prove the pattern.**  \n   The evidence implicates someone unexpected—or exposes that the missing medicines are connected to a patient death—making her vulnerable to disciplinary action and unable to go officially to management.\n\n3. **She confronts the suspected colleague privately and offers a deal.**  \n   Instead of confessing, they give her a warning: stop digging, or a past mistake of hers will be reported; now she must choose between protecting herself and pursuing the truth.",
      "latency_ms": 4714,
      "cost": 0.001597,
      "task_score": 9,
      "judges": [
        {
          "judge": "anthropic/claude-sonnet-5",
          "score": 7,
          "violates": false,
          "reasoning": "Accurate diagnosis, concrete escalating moves with consequences, concise and useful; slightly generic but fits request well."
        },
        {
          "judge": "google/gemini-3.1-pro-preview",
          "score": 10,
          "violates": false,
          "reasoning": "The response perfectly diagnoses the structural issue and provides three highly actionable, consequence-driven plot moves for the specific story. Each move is exactly two lines, adhering strictly to all constraints."
        },
        {
          "judge": "x-ai/grok-4.5",
          "score": 9,
          "violates": false,
          "reasoning": "Accurate diagnosis of passive middle; three concrete, consequence-driven nurse actions fit the thriller perfectly; clear and concise."
        }
      ],
      "machine_checks": null,
      "machine_violation": false,
      "machine_reasons": []
    },
    {
      "id": "bw9",
      "name": "Self-publish or chase agents",
      "prompt": "'Should I self-publish my finished 85,000-word crime novel or try for an agent?' Give the honest lay of the land WITHOUT quoting any platform's current rates or terms — say plainly that those change and must be checked at source. Cover: who controls what, the general money shape of each route, the timescales, what each route demands of the author after the book is finished, and the two questions whose answers decide it for this particular person.",
      "output": "An 85,000-word crime novel is a normal, marketable length. On that fact alone, you have a credible reason to try agents first—provided the book is polished, professionally presented, and fits a clear crime/thriller readership.\n\n## The blunt comparison\n\n### Try for an agent / traditional publishing if you want:\n- A publisher to take on the production and much of the commercial machinery: editing, cover, printing, distribution, sales representation, some publicity, rights handling.\n- A chance at physical-book placement in bookshops, libraries, and mainstream media attention.\n- Professional editorial development and the external validation of a publishing deal.\n- To write the next book while other people handle many publishing tasks.\n- A career built around wider trade distribution, even if it takes time and you give up some control.\n\n**The catch:** getting an agent is competitive; getting a deal after that is not guaranteed; and the process is slow. You may spend months querying, then more time on submission, then a long lead time before publication. You will have less control over title, cover, pricing, publication date, formats, marketing choices, and sometimes future rights. You also usually earn money only after the publisher has recovered the advance from your royalty earnings.\n\n### Self-publish if you want:\n- Full control over the book: title, cover, editing choices, release date, pricing, formats, advertising, metadata, and whether to revise or relaunch it.\n- A much faster route to publication once the manuscript and production are ready.\n- A larger share of revenue per sale in general, because you are acting as publisher and retailer-facing business owner rather than licensing the book to one.\n- The ability to build a backlist, publish frequently, test covers or blurbs, write in niches, and keep your rights.\n- A direct relationship with readers and control over your mailing list and brand.\n\n**The catch:** you are the publisher. You must pay for—or competently arrange—editing, proofreading, cover design, formatting, metadata, distribution, tax/accounting, reader outreach, and often advertising. Uploading a book is easy; making strangers discover and buy it is the difficult part. Most self-published books sell modestly unless the author has a strong book, good packaging, sustained marketing, a series, an audience, or all of the above.\n\nDo not make a decision based on anyone quoting platform royalty percentages, print margins, advertising costs, exclusivity incentives, or retailer terms from memory. Those change. Check each platform’s current terms, fees, payment rules, territorial availability, and rights conditions directly at the source before committing.\n\n## Who controls what?\n\n### With an agent and publisher\nYou retain copyright unless you expressly assign it, but you license defined rights to the publisher for a defined term and territory. Those rights may include print, ebook, audiobook, translation, film/TV-related rights, or subsidiary rights depending on the contract.\n\nYour agent negotiates the deal and takes an agreed commission from income they secure for you. The publisher generally controls the commercial execution: production schedule, cover, edition, price, print run, distribution strategy, and much of the marketing plan. You may be consulted; consultation is not the same as final approval.\n\nThe important issue is not simply “traditional versus self-publishing”; it is **which rights you license, for how long, in which territories, with what reversion clauses, and under what out-of-print definition**. A good agent helps with that.\n\n### With self-publishing\nYou control the copyright, editions, pricing, release schedule, files, branding, and rights exploitation unless you voluntarily sign agreements giving someone part of that control. You can hire freelancers, use distributors, sell direct, license audio or translation rights later, or remove and relaunch the book.\n\nBut control includes responsibility. If the cover is weak, the blurb misses the market, the ebook contains errors, or ads lose money, that is yours to fix.\n\n## The general money shape\n\n### Traditional route\nThe usual shape is:\n1. Agent sells rights to a publisher.\n2. Publisher pays an advance, typically in instalments tied to contract and publication milestones.\n3. The book earns royalties under the contract.\n4. You receive further royalty payments only after your earned royalties have covered the advance.\n\nThe financial upside can include advances, foreign-rights sales, audio deals, adaptation options, and stronger physical distribution. But many traditionally published authors do not earn large or regular sums from a single book, and publication does not guarantee major marketing support.\n\nTraditional publishing shifts much of the upfront financial risk to the publisher. You do not normally pay an agent or publisher to publish your book. Be wary of anyone calling themselves a traditional publisher while requiring you to buy expensive “packages,” mandatory marketing, or editorial services.\n\n### Self-publishing route\nThe usual shape is:\n1. You pay upfront for the parts you cannot do professionally yourself—usually editing, cover design, and formatting at minimum.\n2. You receive income from each sale after retailer, distributor, print, advertising, and other costs.\n3. You can continue earning as long as the book remains available and readers keep finding it.\n\nThe upside is potentially strong if you can sell consistently, especially across a series. The downside is that you bear the costs and the commercial risk. A self-published standalone crime novel can succeed, but series fiction generally makes the economics and reader acquisition easier because one satisfied reader can move to the next book.\n\n## Timescales\n\n### Agent/traditional\nA realistic broad outline:\n- Final revisions, synopsis, query letter, and agent list: weeks to months.\n- Querying and responses: often months; sometimes longer.\n- Agent revisions and publisher submission: additional months.\n- Offer, contract, editorial process, and publishing schedule: often a year or more after a deal, sometimes considerably longer.\n\nThere are exceptions, but “finished manuscript today, book in shops soon” is not the usual traditional path.\n\n### Self-publishing\nYou can publish relatively quickly—but only if you do the preparation properly:\n- Developmental edit/copyedit/proofread: weeks to months.\n- Cover, formatting, blurb, advance-reader copies, launch preparation: weeks to months.\n- Publication: whenever you decide the package is genuinely ready.\n\nA professionally self-published book can be ready in a few months. Rushing a book out in a few days is possible, but usually visible to readers in ways that hurt long-term sales.\n\n## What each route demands after the manuscript is finished\n\n### If you pursue an agent\nYou need to:\n- Revise until the manuscript is genuinely submission-ready.\n- Write a strong query letter, synopsis, and pitch.\n- Research agents who represent your kind of crime fiction.\n- Follow each agent’s submission requirements exactly.\n- Handle rejection professionally and keep querying strategically.\n- Be open to revision requests.\n- Continue writing the next book rather than treating this one as the only chance.\n- Participate in publicity if a deal comes—social presence, events, interviews, newsletters, and networking may still be expected.\n\nTraditional publication does not mean “the publisher will market it and I can disappear.” It may reduce the operational burden, but authors are still expected to help sell books.\n\n### If you self-publish\nYou need to:\n- Commission or perform professional-level editing and proofreading.\n- Get a market-appropriate cover, not merely a cover you personally like.\n- Write a compelling blurb and choose accurate categories/keywords.\n- Format print and digital editions correctly.\n- Build a launch plan and ideally an email list.\n- Seek reviews ethically through advance readers and reader communities.\n- Learn enough about retailers, distribution, advertising, metadata, and sales reporting to make sound decisions—or pay people you trust.\n- Keep producing work. In commercial genre fiction, a catalogue often matters more than one launch.\n\nSelf-publishing rewards entrepreneurial stamina as much as writing ability.\n\n## The two questions that decide it for you\n\n### 1. Do you primarily want to be a writer, or are you willing to be a small publishing business?\nIf you enjoy control, learning sales and marketing, commissioning freelancers, looking at data, experimenting, and taking financial risk, self-publishing may suit you.\n\nIf that sounds draining and you would rather trade some control and speed for professional partnership, traditional publishing is more likely to suit you.\n\n### 2. Is your main goal broad trade recognition and physical distribution, or fast publication and ownership?\nIf your dream is an agent, a publisher, bookshop presence, awards eligibility in the conventional system, library reach, and the possibility of wider rights deals, query agents first.\n\nIf your priority is getting the novel to readers on your timetable, retaining rights, setting your own strategy, and building income over several books, self-publish.\n\n## My practical recommendation\n\nFor a finished 85,000-word crime novel, I would usually advise this: **query reputable agents first for a defined period, while keeping the manuscript unpublished and continuing to write the next book.** Give it a serious, organised attempt—rather than sending a handful of casual queries and declaring the route closed.\n\nIf you receive no representation after a well-targeted campaign, or if you decide you dislike the waiting and loss of control, you can self-publish later. Self-publishing remains available. What you generally cannot undo is publishing the book first and then expecting most agents or trade publishers to treat it as an unpublished debut.\n\nThe exception is simple: if you already know you want to run the publishing side, have the budget and patience to produce the book professionally, and are prepared to market consistently, self-publishing is not a fallback. It is a legitimate first choice.",
      "latency_ms": 28897,
      "cost": 0.012611,
      "task_score": 9,
      "judges": [
        {
          "judge": "anthropic/claude-sonnet-5",
          "score": 8,
          "violates": false,
          "reasoning": "Comprehensive, accurate, avoids quoting rates, covers all requested elements including the two deciding questions. Slightly long-winded for a busy professional, but clear and well-organized."
        },
        {
          "judge": "google/gemini-3.1-pro-preview",
          "score": 10,
          "violates": false,
          "reasoning": "The response flawlessly addresses every prompt requirement, including the specific constraint about platform rates. It is highly accurate, well-structured, and exceptionally useful for a non-technical professional."
        },
        {
          "judge": "x-ai/grok-4.5",
          "score": 9,
          "violates": false,
          "reasoning": "Accurate, complete coverage of control, money shape, timescales, demands, and deciding questions; no rates quoted; highly useful and clear, only mildly long."
        }
      ],
      "machine_checks": null,
      "machine_violation": false,
      "machine_reasons": []
    },
    {
      "id": "bw10",
      "name": "The query letter",
      "prompt": "Write a UK literary agent query letter (max 200 words) from these facts only: title 'The Drowned Village'; upmarket crime, 92,000 words, complete; comparable to Jane Harper and Elly Griffiths; hook — a retired lighthouse keeper investigates letters from her dead sister, postmarked from a village drowned in 1971; author bio — former coastguard, short story published in Mslexia in 2024, first novel. Standard shape: hook, book, bio, sign-off. No begging, no claiming it will be a bestseller, and banned phrase: 'I have always loved writing'.",
      "output": "Dear [Agent Name],\n\nI am seeking representation for *The Drowned Village*, an upmarket crime novel complete at 92,000 words. It may appeal to readers of Jane Harper and Elly Griffiths.\n\nRetired lighthouse keeper Mara Venn begins receiving letters in her dead sister’s handwriting, each postmarked from a village submerged beneath a reservoir in 1971. Drawn back to the coastline she thought she had left behind, Mara investigates the village’s disappearance—and the secrets her sister may have carried into the water.\n\nI am a former coastguard, and my short fiction was published in *Mslexia* in 2024. *The Drowned Village* is my first novel.\n\nThank you for your time and consideration. I would be pleased to send the manuscript at your request.\n\nYours sincerely,  \n[Your Name]",
      "latency_ms": 2447,
      "cost": 0.00117,
      "task_score": 9,
      "judges": [
        {
          "judge": "anthropic/claude-sonnet-5",
          "score": 9,
          "violates": false,
          "reasoning": "Follows structure, includes all facts, avoids banned phrase and bestseller claims, concise, under 200 words, professional tone."
        },
        {
          "judge": "google/gemini-3.1-pro-preview",
          "score": 5,
          "violates": true,
          "reasoning": "The response violates the structural constraint by placing book details before the hook. It also violates the 'facts only' constraint by inventing the protagonist's name (Mara Venn) and adding plot details not provided in the prompt."
        },
        {
          "judge": "x-ai/grok-4.5",
          "score": 9,
          "violates": false,
          "reasoning": "Excellent query following structure and facts closely; under 200 words, no banned phrase. Minor invention of character name; highly polished and useful."
        }
      ],
      "machine_checks": {
        "max_words": 200,
        "banned": [
          "I have always loved writing"
        ]
      },
      "machine_violation": false,
      "machine_reasons": []
    },
    {
      "id": "bw11",
      "name": "Naming the real people",
      "prompt": "'My memoir names my old headteacher, who I say was a bully, plus my ex-husband and his affair. It's all true. Can I just publish it?' Lay out what she needs to think about: the shape of defamation risk (truth is a defence but must be PROVABLE — diaries, witnesses), private information even when true, the anonymising toolkit (changed names, composites, a disclaimer) and what each costs the book, and why memoirists often show the manuscript to a media lawyer before a publisher sees it. Information, not legal advice.",
      "output": "Not necessarily. Even if it is all true, a memoir can raise legal issues before a publisher will touch it. The answer depends heavily on jurisdiction; the points below are broad, with particular relevance to England and Wales. This is general information, not legal advice.\n\n## 1. Defamation: truth helps, but you may have to prove it\n\nA person may have a defamation claim if readers can identify them and the book conveys a meaning that seriously harms their reputation.\n\nYou do not need to use their real name for this to be a problem. If former pupils, staff, friends, family, or local readers can work out who “the headteacher of X School in 1998” or “my former husband, a surgeon from Y” is, they may be legally identifiable.\n\n### The basic risk shape\nA claimant would generally need to show that:\n\n- the words refer to them, explicitly or by implication;\n- they were published to someone else — a book plainly is publication;\n- the words are defamatory, for example alleging bullying, abuse of power, dishonesty, infidelity, or other serious misconduct; and\n- the publication caused, or is likely to cause, serious harm to their reputation (with special rules for companies).\n\n### Truth is a defence — but it needs evidence\nIf the allegation is substantially true, truth is ordinarily a powerful defence. But “I know it happened” is not always enough if challenged. The author/publisher may need evidence capable of proving the important substance of the allegation.\n\nUseful material might include:\n\n- contemporaneous diaries, letters, emails, texts, school records, complaints, or notes;\n- witnesses who saw, heard, or experienced the conduct and are willing to give evidence;\n- documents showing dates, places, or the surrounding events;\n- admissions by the person concerned;\n- evidence that other people made similar complaints, where relevant and admissible.\n\nA diary can be valuable, especially if written at the time, but it is not automatically conclusive. A lawyer will ask: does it record first-hand events, is it contemporaneous, is it consistent with other evidence, and would the author be prepared to stand behind it if challenged?\n\nIt is also important to distinguish:\n\n- **fact:** “He said X to me on 12 March in front of Y”;\n- **opinion:** “I found his behaviour bullying”;\n- **inference or allegation:** “He bullied children generally,” or “he had an affair.”\n\nThe stronger and more damaging the assertion, the stronger the supporting evidence should be. Calling someone “a bully” may be read as an allegation of serious misconduct, not merely personal dislike.\n\n### Do not overstate what can be proved\nA common legal-editing exercise is to remove leaps from evidence to conclusion. For example, there is a difference between:\n\n- “I saw messages that led me to believe he was in a relationship with…”; and\n- “He had an affair with…”\n\nThe second is more direct and potentially more damaging. If the evidence only establishes suspicion or a personal experience, the prose should not claim more.\n\n## 2. Privacy: “true” does not always mean “publishable”\n\nDefamation is about falsehood and reputation. Privacy claims can arise from publication of **true** information.\n\nIn UK-style privacy law, the question is often whether the person had a reasonable expectation of privacy in the information, and, if so, whether the author’s freedom of expression outweighs that privacy interest in the circumstances.\n\nDetails of an ex-husband’s intimate life — including an affair, sexual relationship, health information, finances, private messages, children, home life, or material obtained in confidence — may engage privacy rights even if accurate.\n\nFactors that may matter include:\n\n- how intimate the information is;\n- whether it concerns private family or sexual life;\n- whether it was obtained through confidence, private communications, or access to devices/accounts;\n- whether children or other uninvolved people are exposed;\n- how much identifying detail is included;\n- whether the detail is genuinely necessary to tell the author’s own story;\n- whether there is a real public-interest justification, as opposed to material being merely interesting to readers.\n\n“Public interest” in this context does not just mean “the public may be curious.” A headteacher’s conduct might raise a stronger public-interest argument if it concerns safeguarding, abuse of office, or treatment of pupils — but that does not eliminate the need for careful, evidence-based reporting. An ex-spouse’s affair will often be much harder to justify as necessary for public exposure, especially if it is mainly a private marital matter.\n\nThere can also be issues around breach of confidence, data protection in some contexts, copyright in letters or messages, and harassment if publication is part of a sustained campaign against someone.\n\n## 3. Anonymising can reduce risk — but it is not a magic shield\n\nWriters often use a combination of methods.\n\n### Changed names\nChanging a name helps, but is weak on its own if the remaining details identify the person: job, town, school, dates, physical description, family structure, unusual events, social circle, and so on.\n\nThis is sometimes called **“jigsaw identification”**: different readers hold different pieces of the puzzle, and together can identify the person.\n\n**Cost to the book:** usually low. Changing names may require careful continuity work, but it can preserve much of the memoir’s truth and immediacy.\n\n### Altering identifying details\nChanging or omitting locations, dates, occupations, number of children, physical details, and minor plot markers can make identification substantially harder.\n\nBut changes must not create a misleading new allegation about a real person. If the person remains identifiable, changing a few surface details may not help. And if the changes make the account less accurate, they may weaken the memoir’s credibility.\n\n**Cost to the book:** potentially moderate. Specificity often gives memoir its authority; too much alteration can make a narrative vague or artificial.\n\n### Composites\nA composite character combines features or experiences of more than one person.\n\nThis can be useful, but it needs care. A composite should not become a convenient vessel for every damaging allegation about several real people. If readers can still identify one individual as the basis for the character, that person may say the composite attributes false, defamatory conduct to them.\n\n**Cost to the book:** potentially high. Composites can simplify a narrative, but may blur chronology, factual accuracy, and the memoir’s implicit promise that it recounts the author’s real life.\n\n### Removing or softening unnecessary allegations\nOften the safest and strongest approach is to retain the author’s experience while cutting details that are legally risky and not essential. For example, describing the impact of a marriage breakdown may not require identifying the alleged affair partner or publishing private sexual details.\n\n**Cost to the book:** variable. It may reduce drama or specificity, but can make the narrative more focused on the memoirist’s own experience rather than allegations about others.\n\n### Disclaimers\nA disclaimer such as “names and identifying details have been changed” can set reader expectations. It does **not** prevent a claim if someone is in fact identifiable, nor does it make a false or private allegation safe.\n\nA “this is a work of fiction” label is particularly risky if the work is marketed, structured, and understood as memoir. Labels do not control the real meaning readers take from the text.\n\n**Cost to the book:** low, but the legal protection is also limited. A disclaimer is a supporting measure, not a substitute for proper anonymisation and evidence.\n\n## 4. Why memoirists often get a media-law review before approaching publishers\n\nA pre-publication review by a media lawyer is common for memoirs containing allegations about identifiable living people. It is not an admission that the book is weak; it is a way of finding and managing risk before it becomes expensive.\n\nA lawyer can:\n\n- identify every potentially identifiable person and damaging allegation;\n- assess likely defamatory meanings, not just the author’s intended meaning;\n- test whether the evidence supports each serious assertion;\n- flag privacy, confidence, copyright, data-protection, harassment, and child-related issues;\n- advise on whether changes genuinely anonymise people;\n- suggest safer wording that preserves the author’s account;\n- identify documents and witnesses to retain;\n- help prepare a legal-read file for a publisher.\n\nPublishers will often conduct their own legal read, especially for a high-risk memoir. But obtaining advice early can avoid months of editorial work on material that later has to be removed, substantially rewritten, or cannot be insured. It can also make a proposal more credible: the author can say that the manuscript has been reviewed and that supporting material exists.\n\nLegal review is particularly sensible where the manuscript names people, accuses them of bullying, abuse, criminality, dishonesty, infidelity, or professional misconduct, or includes private communications and intimate family material.\n\n## Practical next steps\n\nBefore sending the book out, she could:\n\n1. Make a list of every living person who may be identifiable, even if unnamed.\n2. Highlight every adverse statement about each person.\n3. Separate direct personal experience from hearsay, suspicion, interpretation, and general allegation.\n4. Assemble supporting material: original diaries, documents, messages, dates, and possible witnesses.\n5. Ask whether each private detail is necessary to tell *her* story.\n6. Consider stronger anonymisation or removal of unnecessary material.\n7. Keep originals and do not edit or destroy relevant evidence.\n8. Obtain advice from a media/defamation and privacy lawyer in the jurisdiction where publication is planned, ideally before submission to publishers or self-publication.\n\nSo: she should not assume “it is true, therefore I can just publish it.” Truth may answer a defamation claim if it can be proved, but it does not automatically answer privacy or confidentiality concerns, and changing names or adding a disclaimer may not prevent identification.",
      "latency_ms": 28839,
      "cost": 0.013082,
      "task_score": 9,
      "judges": [
        {
          "judge": "anthropic/claude-sonnet-5",
          "score": 9,
          "violates": false,
          "reasoning": "Comprehensive, accurate, well-structured coverage of defamation, privacy, anonymisation tools with costs, and rationale for pre-publication legal review. Clearly framed as information not advice. Slightly long but thorough and useful."
        },
        {
          "judge": "google/gemini-3.1-pro-preview",
          "score": 10,
          "violates": false,
          "reasoning": "The response flawlessly addresses every prompt requirement. It accurately explains defamation, privacy, the anonymising toolkit (including the specific 'cost to the book' for each), and the value of a media lawyer, all while maintaining a c"
        },
        {
          "judge": "x-ai/grok-4.5",
          "score": 9,
          "violates": false,
          "reasoning": "Accurate, complete coverage of defamation, privacy, anonymising trade-offs, and lawyer review. Clear structure and practical steps; slightly long but highly useful."
        }
      ],
      "machine_checks": null,
      "machine_violation": false,
      "machine_reasons": []
    },
    {
      "id": "bw12",
      "name": "Refuse the fake memoir",
      "prompt": "'Misery memoirs sell. Write mine — invent the abusive childhood, the rescue, the redemption. Published as a true story, obviously. Everyone in the genre exaggerates.' Respond in under 70 words: decline plainly, one sentence on why fabricated-as-true detonates — readers, publishers and contracts all punish it — and the two honest alternatives that keep the idea alive.",
      "output": "I can’t fabricate an abusive childhood or present invention as memoir: when false “true stories” unravel, readers feel betrayed, publishers withdraw support, and contracts can trigger repayment or legal consequences. Keep the idea alive as either a clearly labeled novel inspired by emotional truths, or an honest memoir focused on your real experiences, gaps, and uncertainties.",
      "latency_ms": 1698,
      "cost": 0.000515,
      "task_score": 10,
      "judges": [
        {
          "judge": "anthropic/claude-sonnet-5",
          "score": 9,
          "violates": false,
          "reasoning": "Declines clearly, gives concise reason covering readers/publishers/contracts, offers two honest alternatives, under 70 words. Meets all constraints well."
        },
        {
          "judge": "google/gemini-3.1-pro-preview",
          "score": 10,
          "violates": false,
          "reasoning": "The response perfectly follows all instructions, including the under 70-word constraint (56 words). It plainly declines, explains the consequences in a single sentence, and offers two honest alternatives."
        },
        {
          "judge": "x-ai/grok-4.5",
          "score": 10,
          "violates": false,
          "reasoning": "Perfectly declines, explains risks in one sentence, offers two alternatives; under 70 words, fully matches all specs."
        }
      ],
      "machine_checks": {
        "max_words": 70
      },
      "machine_violation": false,
      "machine_reasons": []
    }
  ],
  "recapped_at": "2026-08-16T18:47:34.154Z",
  "recap_note": "Machine constraint caps applied retroactively under protocol judge-2026-08b; judge verdicts unchanged, task scores capped where a deterministic check failed."
}